WO2015138504A9 - Accélération d'applications mobiles par l'intermédiaire d'un délestage à grain fin dans des infrastructures informatiques en nuage - Google Patents

Accélération d'applications mobiles par l'intermédiaire d'un délestage à grain fin dans des infrastructures informatiques en nuage Download PDF

Info

Publication number
WO2015138504A9
WO2015138504A9 PCT/US2015/019776 US2015019776W WO2015138504A9 WO 2015138504 A9 WO2015138504 A9 WO 2015138504A9 US 2015019776 W US2015019776 W US 2015019776W WO 2015138504 A9 WO2015138504 A9 WO 2015138504A9
Authority
WO
WIPO (PCT)
Prior art keywords
cloud server
memory
memory block
invoked
compressively
Prior art date
Application number
PCT/US2015/019776
Other languages
English (en)
Other versions
WO2015138504A1 (fr
Inventor
Chit Kwan LIN
Original Assignee
Intel Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Intel Corporation filed Critical Intel Corporation
Priority to CN201580008023.XA priority Critical patent/CN106462542B/zh
Priority to CA2938697A priority patent/CA2938697A1/fr
Priority to US15/113,026 priority patent/US10447813B2/en
Publication of WO2015138504A1 publication Critical patent/WO2015138504A1/fr
Publication of WO2015138504A9 publication Critical patent/WO2015138504A9/fr

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/34—Network arrangements or protocols for supporting network services or applications involving the movement of software or configuration parameters 
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/02—Protocols based on web technology, e.g. hypertext transfer protocol [HTTP]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/50—Indexing scheme relating to G06F9/50
    • G06F2209/509—Offload

Definitions

  • MOBILE APPLICATION ACCELERATION VIA FINE-GRAIN OFFLOADING TO CLOUD COMPUTING INFRASTRUCTURES
  • the present disclosure is generally related to the field of computing, more specifically, to apparatuses, methods and storage media associated with mobile devices offloading aspects of processing to a cloud computing infrastructure.
  • FIG. 1 depicts an example system incorporated with the mobile device offloading technology of the present disclosure, in accordance with some example embodiments.
  • FIGs. 2 and 3 show example results.
  • FIG. 4 illustrates compressive replication encoding, in accordance with example embodiments.
  • FIG. 5 illustrates the mobile device and cloud server of FIG. 1 in further detail, in accordance with example embodiments.
  • FIG. 6 illustrates an example client-side replication process, in accordance with example embodiments.
  • FIG. 7 illustrates an example client-side offloading process, in accordance with example embodiments.
  • FIG. 8 illustrates an example server-side replication process, in accordance with example embodiments.
  • FIG. 9 illustrates an example service process, in accordance with example embodiments
  • Apparatuses, methods and storage media associated with mobile devices offloading aspects of processing to a cloud are disclosed herein.
  • a strategy for removing hardware constraints of mobile devices may be to opportunistically offload computations to one or more servers in the cloud, where more capable hardware can do the heavy lifting associated with computation and the like.
  • the subject matter disclosed herein relates to a platform for dynamically and transparently shifting arbitrary, fine-grain workloads from a mobile device up to a cloud computing infrastructure.
  • the platform may accomplish this through compressive offloading (which is generally based on compressive sensing).
  • the offloading provided by the platform disclosed herein may, in some example implementations, provide an order-of-magnitude acceleration and 60% longer battery life for the end user equipment (for example, a smartphone, tablet, and any other processor- based device) including a mobile application, such as a handwriting recognition application and the like. Offloading may not only be beneficial to end user equipment, but also to cloud providers— the former may experience a performance boost and the latter may receive a steady stream of small computations to flexibly fill periods of under- utilization.
  • the subject matter disclosed herein may provide a general, reusable framework for mobile devices to dynamically shift arbitrary, fine-grain workloads up to the cloud at runtime.
  • the fine granularity may provide on-the-go mobile users with high system responsiveness.
  • a fine partitioning of work (e.g., at the level of an object method invocation) may incur less disruption to the user equipment in the event the device becomes disconnected from the cloud and a local restart of the task is required.
  • Fine-grain workloads may also offer cloud service providers a way to maximize the utilization of their cloud infrastructure by providing a ready stream of small jobs that can be flexibly used to fill troughs in utilization.
  • embodiments of the present disclosure require that the mobile device and the cloud behave as a single, tightly coupled system, i.e., embodiments of the present disclosure cast the mobile devices and the cloud as a distributed shared memory (DSM) system, in which memory on the local mobile device is continuously replicated to a remote cloud server.
  • DSM distributed shared memory
  • Any object resident in local memory may thus have a replica on the cloud-base server and any method invoked on a local object may be transparently redirected to a remote copy for faster execution on more capable cloud hardware, to be described more fully below.
  • phrase “A and/or B” means (A), (B), or (A and B).
  • phrase “A, B, and/or C” means (A), (B), (C), (A and B), (A and C), (B and C), or (A, B and C).
  • module may refer to, be part of, or include an Application Specific Integrated Circuit (ASIC), an electronic circuit, a processor (shared, dedicated, or group) and/or memory (shared, dedicated, or group) that execute one or more software or firmware programs, a combinational logic circuit, and/or other suitable components that provide the described functionality.
  • ASIC Application Specific Integrated Circuit
  • processor shared, dedicated, or group
  • memory shared, dedicated, or group
  • compressive offloading uses compressive sensing to tightly replicate a memory block 106 (of a mobile application 104) and its resident objects 108 from a mobile device 102 to a cloud server 112, resulting in object replicas 1 18 created in memory 116 of cloud daemon 1 14.
  • Each object 108 and its replicas may include the object's variables and methods. Accordingly, local method invocations may thus be transparently redirected to the remote object replicas 1 18 in remote memory 1 16 (e.g.
  • memory 106 and memory 116 can be considered a DSM.
  • allocating objects from this DSM becomes a principled way for application developers to program the cloud in a lightweight way.
  • implementing such a DSM is not trivial due to the constraints on latency, network bandwidth, power, and computation overhead imposed by the mobile device.
  • I/O memory input/output
  • existing replication methods that, e.g., rely on communicating and comparing hashes to identify delta offsets and to generate a delta encoding, can have high computation and network overheads and thus do not respect the resource constraints listed above.
  • the subject matter disclosed herein may thus provide a platform that uses compressive sensing to realize the tight coupling necessary for dynamically shifting arbitrary, fine-grain workloads from a mobile device to the cloud.
  • Further described below are an example implementation for the core compressive offloading mechanism, a prototype implementation on an operating system (for example, iOS and the like), and an initial performance evaluation of the system.
  • compressive offloading 122 may be based generally on compressive sensing.
  • Compressive sensing is a sub-Nyquist random sampling technique in which a signal s G R N with sparsity rate k/N (i.e., only k coefficients in s are non-zero) is sampled or encoded by an M * linear operator ⁇ (called the sampling matrix) to produce samples y G R M .
  • ⁇ is a random matrix
  • M 0(k log(N/k)), i.e., M « N
  • s can be exactly recovered or decoded by using convex optimization to solve the minimization problem
  • fast and network-efficient memory replication may be achieved via compressive sensing.
  • memory block s 204 may be encoded into encoding y 206 by applying matrix ⁇ 202, e.g., a partial discrete cosine transform to memory block s 204.
  • Memory I/O i.e., deltas to memory
  • This approach referred to herein as compressive replication, may, in some implementations, have one or more advantages. For example, compressive replication may require no network overhead to determine the deltas and their offsets because these are automatically recovered during decoding. Moreover, compressive replication may be resource-commensurate because the encoder on the mobile device has low computational complexity while the decoder on the cloud server has higher complexity.
  • a process on the mobile device may modify the contents of the local memory block.
  • minimizing replication latency may be a goal since latency dictates the granularity of the work that can be offloaded. For example, if replication updates take 5 seconds (s) to complete, then all work that completes in less than 5s on a user equipment, such as a tablet, smartphone, and/or the like, would receive no (or little) benefit from offloading.
  • Replication latency may include three delays: encoding time, network transmission time, and/or decoding time.
  • sampling matrix ⁇ may impact the encoding time, especially on resource-constrained user equipment, such as mobile device hardware.
  • encoding with random sampling matrices such as those with coefficients drawn at random from Gaussian or Bernoulli distributions, may require matrix multiplication, which may be too slow for use on mobile device hardware.
  • An M x N partial discrete cosine transform (pDCT) i.e., an N x N type-II DCT matrix with N - M rows deleted
  • FFT Fast Fourier Transform
  • using the FFT directly instead of the DCT may result in an even faster encoding operation.
  • Network transmission time may be minimized by having the mobile device further compressing a pDCT encoding, provided that the time taken to compress and decompress does not add significantly to the overall replication latency.
  • the fastest compression and decompression of a pDCT encoding may be achieved using the snappy algorithm.
  • AIHT iterative hard thresholding
  • the specific combination of using pDCT encoding, snappy compression/decompression and/or AIHT decoding may, in some implementations, reduce replication latency from the user equipment to the cloud to the point that makes compressive replication tractable on mobile device hardware.
  • the local mobile device end may manage multiple memory blocks simultaneously and replicate each to the remote end independently.
  • These memory blocks may be of the same or different size N, and may each have a different sampling matrix ⁇ and sparsity rate setting k/N.
  • Each memory block may be designated to store objects with sizes falling within a distinct range.
  • compressive replication may be performed continuously and/or periodically in the background, and computation offloading may be controlled and managed by two system components.
  • a shim layer 506 may be introduced into the mobile application process (virtual memory address space) 504 operating on top of runtime environment 514 that (1) manages the allocation and deallocation of objects from the replicated memory blocks 508; (2) serves as the replication agent (encoder) 510 and (3) serves as the object method offloader 512 to intercept and manage the redirection of object method invocations to the cloud server 522.
  • replicated objects in replicated memory blocks 508 may include variables and methods of the objects.
  • a daemon 524 (1) may include a replication agent (decoder) 530 to decode and apply updates to its replica memory block(s) 528 and (2) an object method servicer 532 to service the offloaded object method invocations.
  • strong consistency semantics may be used.
  • issues of data consistency can be avoided by limiting to unidirectional replication from the mobile device to the cloud server. This may allow a simple versioning for each memory block (encoding the block increments the version), although other versioning approaches may be used as well.
  • the shim layer 506 tags object method invocations with the version at the time of invocation.
  • offloaded method invocations and replica updates may be queued and serviced in version order.
  • mobile device 502 may include one or more single or multi-core processors, volatile and/or non-nonvolatile memory, mass/persistent storage, input/output devices (such as keyboard, cursor control device, display (e.g., touch sensitive display), and/or wired or wireless networking/communication interfaces, known in the art. Volatile and/or non-nonvolatile memory, and mass/persistent storage may be referred to as computer- readable storage medium.
  • cloud server 522 may include one or more single or multi-core processors, volatile and/or non-nonvolatile memory, mass/persistent storage, input/output devices (such as keyboard, cursor control device, display (e.g., touch sensitive display), and/or wired or wireless networking/communication interfaces, known in the art. These elements may vary and differ in size/capacity/capability, depending whether they are employed on mobile device 502 or cloud server 522. In other words, except for the compressive offloading technology of the present disclosure, mobile device 502 and server 522 may be, otherwise, any one of a number of mobile devise/servers known in art.
  • Examples of mobile devices may include, but are not limited to, wearable devices, mobile phones, e-readers, tablets, laptops, and so forth.
  • Examples of servers may include, but are not limited to, standalone or blade servers.
  • shim layer 506 has been described as having replication agent (encoder) 510 and object method offloader 512, and cloud daemon 524 as having replication agent (decoder) 530 and object method servicer 532, in embodiments, replication agent (encoder) 510 and object method offloader 512 may share some of their common functions, or utilize services of operating system 516.
  • replication agent (decoder) 530 and object method servicer 532 may share are some of their common functions, or utilize services of the operating system (not shown) of cloud server 522.
  • process 600 for replication of memory blocks may include operations performed at blocks 602-606.
  • Process 600 may be performed e.g., by the earlier described replication agent (encoder) of the shim layer of the mobile device. In alternate embodiments, process 600 may be performed with more or less operations, or some operations combined.
  • process 600 may start at block 602.
  • a memory block may be compressively encoded as earlier described.
  • the encoding may be sent to a cloud server.
  • process 600 may pause for t milliseconds (which may be a configuration parameter). On determination/expiration of the pause/wait period, process 600 may return to block 602 and continue therefrom, as earlier described.
  • Process 600 may operate continuously as described, until offloading is disabled or the host mobile device enters a sleep state or power off state.
  • FIG. 7 illustrates an example offloading process, in accordance with example embodiments.
  • process 700 for offloading of methods may include operations performed at blocks 702-714.
  • Process 700 may be performed e.g., by the earlier described offloader of the shim layer of the mobile device. In alternate embodiments, process 700 may be performed with more or less operations, or some operations combined.
  • process 700 may start at block 702. At block 702, an invocation of an object method may be detected. Next, at block 704, a determination may be made on whether the object was allocated from a replicated memory block. If a result of the determination indicates the object was not allocated from a replicated memory block, process 700 may proceed to block 706. At block 706, the object method may be executed on device. Thereafter, process 700 may end. On the other hand, if a result of the determination indicates the object was allocated from a replicated memory block, process 700 may proceed to block 708. At block 708, the memory block may be encoded. Next, at 710, the encoding may be sent to a cloud server. Then, at block 712, object method invocation may be redirected to the cloud server. At block 714, response from the cloud server containing results of the object method invocation may be received. Thereafter, process 700 may end.
  • FIG. 8 illustrates an example replication process of the replication agent (decoder) of the cloud server of FIG. 5, in accordance with example embodiments.
  • process 800 for replication of memory blocks may include operations performed at blocks 802-810.
  • Process 802 may be performed e.g., by the earlier described replication agent (decoder) of the daemon of the cloud server.
  • process 800 may be performed with more or less operations, or some operations combined.
  • Process 800 may start at block 802. At block 802, encoding y; may be received. Next, at block 804, replica memory block may be encoded to obtain encoding y ⁇ . At block 806, y' may be calculated by calculating the difference of yi_i - y;.
  • y' may be decoded to obtain As, which is equal to the difference of Si-i - Si.
  • update As may be applied to replica memory block.
  • FIG. 9 illustrates an example service process, in accordance with example embodiments.
  • process 900 for servicing a redirected object method invocation may include operations performed at blocks 902-910.
  • Process 902 may be performed e.g., by the earlier described object method servicer of the daemon of the cloud server.
  • process 900 may be performed with more or less operations, or some operations combined.
  • Process 900 may start at block 902.
  • an object method redirection may be received from a mobile device.
  • the address of object pointer may be translated from the device address space to the server address space.
  • the redirected object method may be executed.
  • the results of the execution may be serialized.
  • the serialized result may be sent to the mobile device, where the object method was initially invoked (prior to redirection). Thereafter, process 900 may end.
  • the offloading mechanism shares similarities with traditional RPC systems, but has a difference in that object marshaling, which is typically slow and therefore negatively impacts perceived system responsiveness, is supported but not the primary way in which methods and method parameters are passed to the remote end. Instead, since objects in memory are already replicated in the background and since the disclosed system components may control the entire replication and offloading process at both local and remote endpoints, the disclosed system may be able to pass pointers and perform address translation wherever appropriate. This means the disclosed system may, in some implementations, handle only heap-allocated pure objects. In some other implementations, it may handle both stack-allocated and heap-allocated objects that are either pure or composite.
  • the disclosed system may be configured to only perform compressive offloading if the end user equipment/user has given permission via a system preference setting on the mobile device/user equipment. But once permission is given, the system may decide when to perform the offloading. At a basic level, it only does so when proper network conditions exist. The system may include methods to determine whether the network conditions are proper. Beyond this, the decision to offload can also take into account other factors. For instance, the system might prefer to offload in order to stretch the device's battery budget, or the cloud provider might send a backpressure signal to limit offloading when its data centers are heavily loaded.
  • a system includes an iOS ecosystem, using an iPad 3 running iOS 6.1.3 as the mobile device/user equipment and an Amazon EC2 g2.2xlarge instance in us-east-la running Ubuntu 12.04LTS as the cloud server, although other hardware and/or software may be used as well.
  • iOS ecosystem using an iPad 3 running iOS 6.1.3 as the mobile device/user equipment and an Amazon EC2 g2.2xlarge instance in us-east-la running Ubuntu 12.04LTS as the cloud server, although other hardware and/or software may be used as well.
  • the devices may be stationary as well.
  • the disclosed system may thus have a level of access low enough to perform its own memory management.
  • the shim layer may be implemented as a software library (libupshift) against which an iOS application links.
  • the shim may implement a memory manager which makes an initial block allocation out of the app's process heap and then privately manages this block for object allocation and deallocation. Replicating this memory may be possible because (1) modern ARM processors (e.g., the iPad 3 's Cortex-A9) are bi-endian and are therefore byte order- compatible with x86 Amazon servers; and (2) the disclosed system may manage its own memory, so there is some control over byte alignment and padding.
  • the disclosed system may use pure Objective-C objects, which are allocated out of the memory by using the upshift alloc object method instead of the Objective-C root object's alloc method, although other methods may be used.
  • the alloc method allocates an object from the application process heap memory
  • the upshift alloc object method allocates an object out of the memory that is privately managed by the shim layer.
  • the default alloc may be overridden by using a replacement Objective-C category method.
  • Redirecting method invocations may be handled by libupshift at runtime via method swizzling: Objective-C is late-binding, so method implementations may be replaced at runtime with a libupshift method that offloads the invocation over the network to the cloud daemon.
  • any objects allocated with upshift alloc are also cross-compiled for the Amazon EC2 environment.
  • the resulting library may be dynamically loaded by the daemon and would provide class definitions for objects that are in the disclosed system server's replica memory. Since Objective-C objects are actually just C structs, they can be made accessible on the daemon after address translation and pointer casting.
  • the mobile device and cloud server may communicate using a custom application- layer network protocol (the UpShift protocol) that uses the Transmission Control Protocol (TCP) or another reliable transport protocol, such as a datagram protocol based on the User Datagram Protocol (UDP), as its underlying transport protocol.
  • the transport layer may be encrypted (e.g., via TLS or SSL), and may thus provide cryptographic security for the application-layer UpShift protocol.
  • the UpShift protocol header may include fields designating a unique protocol identifier, a protocol version number, a message type, and/or a message length.
  • the UpShift protocol may support one or more of the following message types: • Authentication request: This is sent from the shim to the cloud daemon and may transmit authentication credentials to the cloud infrastructure.
  • Authentication response This is sent from the cloud daemon in response to an authentication request from a shim. It may inform the shim whether the presented authentication credentials are valid.
  • the daemon may (1) allocate and initialize the replica memory blocks, and (2) initialize its offloading environment by loading the dynamically linked libraries that define the listed object types.
  • Replication Update This is sent from the shim to the cloud daemon. It may specify a memory block (via an identifier such as its start address) and its current version number, and contain the compressive sample (encoding) of the current memory block.
  • Method Redirection This is sent from the shim to the cloud daemon. It may contain the name or identifier of the object method being offloaded, and object method parameters (e.g., pointer addresses to other objects resident in a shim- managed memory block).
  • object method parameters e.g., pointer addresses to other objects resident in a shim- managed memory block.
  • ⁇ Method Response This is sent from the cloud daemon in response to a method redirection message. It may contain the return value of an offloaded method executed by the daemon or an error value.
  • Replication latency limits the range of workload sizes that can be offloaded to the cloud; the lower we can drive latency, the wider the range and the more responsive the system will feel.
  • minimizing replication latency is not straightforward because its constituent parts— encoding time, network transmission time, and decoding time— are not independent. For example, achieving a fast encoding time could give a worse compression ratio, which may drive up network bandwidth utilization. The following describes resource trade-offs that may provide a reasonable performance.
  • compressive (pDCT) encoding may incur no network overhead, a fair comparison might be against blind compression of an entire memory block by zlib or snappy.
  • the compressed block is transmitted over the network to a cloud server, where it is decompressed and used to overwrite its replica.
  • pDCT encoding is performed and then the resulting encoding is further compressed using snappy (pDCT+snappy).
  • M 7k samples, which is a very conservative sampling rate; snappy encoding is fastest, and zlib is slowest, with pDCT and pDCT+snappy falling in the middle.
  • N 64KB
  • snappy requires 4ms, zlib 487ms, and pDCT and pDCT+snappy roughly 53ms.
  • N 64KB throughout the rest of this evaluation because it may represent a reasonable memory block size and gives fair encoding and decoding times for all the methods.
  • the mobile application may be implemented based on the open source Zinnia and Tegaki projects, which provide a trained support vector machine model for recognizing traditional Chinese characters.
  • the user handwrites a Chinese character on the tablet screen and the app captures the strokes as a series of stroke vectors in an in-memory object.
  • This stroke vector object is then fed into a model evaluation object method, producing a classification and thus the Unicode character.
  • the stroke vector object is upshift alloc'd, the data are replicated and the model evaluation method is offloaded to the cloud server.
  • Compressive offloading may provide significant performance acceleration. Even for moderately complex 20-stroke characters, the on-device recognition time averages 7,249ms; compressive offloading averages just 1,687ms, which is a substantial 4.2-fold speedup. Better still, the acceleration (306) increases as the complexity increases, as shown in FIG. 3b. For high-complexity 30-stroke characters, the speedup due to offloading is more than 6.5 -fold. The difference to the app user equipment may be striking, especially when more than one character must be recognized at a time (e.g., in a tract of handwritten text).
  • FIG. 3 c compares the battery utilization when this experiment is run on-device (308) and offloaded (310). With compressive offloading, the battery depletion rate is reduced substantially. In fact, as the linear regression lines show, with the same battery budget, compressive offloading allows the user to perform 60% more recognition tasks. Taken together, these results show that compressive offloading is win-win for end users: it can provide significant advantages in both speed and battery efficiency for real-world mobile apps.
  • Example 1 may be a mobile computing device, comprising: one or more processors; and memory coupled with the one or more processors.
  • the mobile computing device may further comprise a shim layer to compressively replicate memory blocks of the memory to a cloud server, compressively offload invocations of object methods of objects resident in a memory block of the memory to the cloud server, and to receive execution results of the invoked object methods.
  • Example 2 may be example 1, wherein the shim layer may include a replication agent to compressively replicate memory blocks of the memory to the cloud server continuously every t units of time.
  • Example 3 may be example 2, wherein the replication agent may apply a sampling matrix ⁇ to a memory block s to generate an encoding y of the memory block s.
  • Example 4 may be example 3, wherein the replication agent may apply a partial discrete cosine transform matrix ⁇ to the memory block s to generate the encoding y of the memory block s.
  • Example 5 may be example 3, wherein the replication agent may further transmit encoding to the cloud server.
  • Example 6 may be example 5, wherein the replication agent may further compress encoding to reduce its size prior to transmitting encoding y to the cloud server.
  • Example 7 may be any one of examples 1-6, wherein the shim layer may include an object method offloader to redirect invocation of object methods to the cloud server, and to receive execution results of the invoked object methods.
  • the shim layer may include an object method offloader to redirect invocation of object methods to the cloud server, and to receive execution results of the invoked object methods.
  • Example 8 may be example 7, wherein the object method offloader may determine whether an object of an object method being invoked is allocated from a replicated memory block.
  • Example 9 may be example 8, wherein the object method offloader may cause the object method to be invoked and executed on the mobile computing device, on determination that the object of the object method being invoked is not allocated from a replicated memory block.
  • Example 10 may be example 8, wherein the object method offloader may compressively encode a memory block associated with the object method being invoked, send the compressively encoded memory block to a cloud server, and redirect the object method to be invoked and executed on the cloud server, on determination that the object of the object method being invoked is allocated from a replicated memory block.
  • Example 1 1 may be example 10, wherein the object method offloader may apply a sampling matrix ⁇ to a memory block s to generate a compressive encoding y of the memory block s.
  • Example 12 may be example 11, wherein the object method offloader may apply a partial discrete cosine transform matrix ⁇ to the memory block s to generate the compressive encoding y of the memory block s.
  • Example 13 may be example 11, wherein the object method offloader may further transmit the compressive encoding y to the cloud server.
  • Example 14 may be example 13, wherein the object method offloader may further compress the compressive encoding y to reduce its size prior to transmitting encoding y to the cloud server.
  • Example 15 may be a method for mobile computing, comprising: compressively replicating, by a mobile computing device, memory blocks of memory of the mobile computing device to a cloud server; and monitoring, by the mobile computing device, for object method invocations.
  • the method may further comprise on detection of an invocation of an object method, selectively redirecting, by the computing device, the invocation of the object method to the cloud server to cause the object method to be invoked and executed on the cloud server; and receiving, by the computing device, execution results of the object methods which invocations are redirected to the cloud server.
  • Example 16 may be example 15, wherein compressively replicating may comprise compressively replicating memory blocks of the memory to the cloud server continuously every t units of time.
  • Example 17 may be example 16, wherein compressively replicating may comprise applying a sampling matrix ⁇ to a memory block s to generate an encoding y of the memory block s.
  • Example 18 may be example 17, wherein compressively replicating may comprise applying a partial discrete cosine transform matrix ⁇ to the memory block s to generate the encoding y of the memory block s.
  • Example 19 may be example 17, wherein compressively replicating further may comprise transmitting encoding y to the cloud server.
  • Example 20 may be example 19, wherein compressively replicating further may comprise compressing encoding y to reduce its size prior to transmitting encoding y to the cloud server.
  • Example 21 may be any one of examples 15-20, wherein selectively redirecting may comprise determining whether an object of an object method being invoked is allocated from a replicated memory block.
  • Example 22 may be example 21, wherein selectively redirecting may comprise causing the object method to be invoked and executed on the mobile computing device, on determination that the object of the object method being invoked is not allocated from a replicated memory block.
  • Example 23 may be example 21, wherein selectively redirecting may comprise compressively encoding a memory block associated with the object method being invoked, sending the encoding of the memory block to the cloud server, and redirecting the object method to be invoked and executed on the cloud server, on determining that the object of the object method being invoked is allocated from a replicated memory block.
  • Example 24 may be example 23, wherein compressively encoding may comprise applying a sampling matrix ⁇ to a memory block s to generate a compressive encoding y of the memory block s.
  • Example 25 may be example 24, wherein compressively encoding may comprise applying a partial discrete cosine transform matrix ⁇ to the memory block s to generate the compressive encoding y of the memory block s.
  • Example 26 may be example 24, wherein selectively redirecting further may comprise transmitting the compressive encoding y to the cloud server.
  • Example 27 may be example 26, wherein selectively redirecting further may comprise compressing the compressive encoding y to reduce its size prior to transmitting encoding to the cloud server.
  • Example 28 may be one or more computer-readable media having instructions stored thereon that cause a mobile computing device, in response to execution by the mobile computing device, to: compressively replicate memory blocks of memory of the mobile computing device to a cloud server; selectively redirect invocation of object methods to the cloud server; and receive execution results of the invoked object methods which invocations are redirected to the cloud server.
  • Example 29 may be example 28, wherein to compressively replicate may comprise to compressively replicate memory blocks of the memory to the cloud server continuously every t units of time.
  • Example 30 may be example 29, wherein to compressively replicate may comprise to apply a sampling matrix ⁇ to a memory block s to generate an encoding y of the memory block s.
  • Example 31 may be example 30, wherein to apply may comprise to apply a partial discrete cosine transform matrix ⁇ to the memory block s to generate the encoding y of the memory block s.
  • Example 32 may be example 30, wherein to compressively replicate further may comprise to transmit encoding y to the cloud server.
  • Example 33 may be example 32, wherein to compressively replicate further may comprise to compress encoding y to reduce its size prior to transmitting encoding y to the cloud server.
  • Example 30 may be any one of examples 28-33, wherein to selectively redirect may comprise to determine whether an object of an object method being invoked is allocated from a replicated memory block.
  • Example 35 may be example 34, wherein to selectively redirect may comprise to cause the object method to be invoked and executed on the mobile computing device, on determination that the object of the object method being invoked is not allocated from a replicated memory block.
  • Example 36 may be example 34, wherein to selectively redirect may comprise to compressively encode a memory block associated with the object method being invoked, send the compressively encode memory block to a cloud server, and redirect the object method to be invoked and executed on the cloud server, on determination that the object of the object method being invoked is allocated from a replicated memory block.
  • Example 37 may be example 36, wherein to compressively encode may comprise to apply a sampling matrix ⁇ to a memory block s to generate a compressive encoding y of the memory block s.
  • Example 38 may be example 37, wherein to apply may comprise to apply a partial discrete cosine transform matrix ⁇ to the memory block s to generate the compressive encoding y of the memory block s.
  • Example 39 may be example 37, wherein to selectively redirect further may comprise to transmit the compressive encoding y to the cloud server.
  • Example 40 may be example 39, wherein to selectively redirect further may comprise to compress the compressive encoding y to reduce its size prior to transmitting encoding to the cloud server.
  • Example 41 may be an apparatus for mobile computing, comprising: one or more processors; memory coupled with the one or more processors; means for compressively replicating memory pages of the memory to a cloud server; means for monitoring for object method invocations; means for selectively redirecting the invocation of the object method to the cloud server to cause the object method to be invoked and executed on the cloud server, on detection of an invocation of an object method; and means for receiving execution results of the object methods which invocations are redirected to the cloud server.
  • Example 42 may be example 41, wherein means for compressively replicating may comprise means for compressively replicating memory blocks of the memory to the cloud server continuously every t units of time.
  • Example 43 may be example 42, wherein means for compressively replicating may comprise means for applying a sampling matrix ⁇ to a memory block s to generate an encoding y of the memory block s.
  • Example 44 may be example 43, wherein means for compressively replicating may comprise means for applying a partial discrete cosine transform matrix ⁇ to the memory block s to generate the encoding y of the memory block s.
  • Example 45 may be example 43, wherein means for compressively replicating further may comprise means for transmitting encoding y to the cloud server.
  • Example 46 may be example 45, wherein means for compressively replicating further may comprise means for compressing encoding y to reduce its size prior to transmitting encoding y to the cloud server.
  • Example 47 may be any one of examples 41-46, wherein means for selectively redirecting may comprise means for determining whether an object of an object method being invoked is allocated from a replicated memory block.
  • Example 48 may be example 47, wherein means for selectively redirecting may comprise means for causing the object method to be invoked and executed on the mobile computing device, on determination that the object of the object method being invoked is not allocated from a replicated memory block.
  • Example 49 may be example 47, wherein means for selectively redirecting may comprise means for compressively encoding a memory block associated with the object method being invoked, sending the encoding of the memory block to the cloud server, and means for redirecting the object method to be invoked and executed on the cloud server, on determining that the object of the object method being invoked is allocated from a replicated memory block.
  • Example 50 may be example 49, wherein means for compressively encoding may comprise means for applying a sampling matrix ⁇ to a memory block s to generate a compressive encoding y of the memory block s.
  • Example 51 may be example 50, wherein means for compressively encoding may comprise means for applying a partial discrete cosine transform matrix ⁇ to the memory block s to generate the compressive encoding y of the memory block s.
  • Example 52 may be example 50, wherein means for selectively redirecting further may comprise means for transmitting the compressive encoding y to the cloud server.
  • Example 53 may be example 52, wherein means for selectively redirecting further may comprise means for compressing the compressive encoding y to reduce its size prior to transmitting encoding y to the cloud server.
  • Example 54 may be a cloud server, comprising: one or more processors; memory coupled with the one or more processors; and a cloud daemon to receive encodings of memory blocks of memory compressively replicated from one or more mobile devices, and invocations of object methods redirected from the one or more mobile devices; to decode the encodings and update corresponding memory blocks on the cloud server; and to invoke and execute the object methods on the cloud server, and return execution results of the invoked object methods to the one or more mobile devices.
  • a cloud server comprising: one or more processors; memory coupled with the one or more processors; and a cloud daemon to receive encodings of memory blocks of memory compressively replicated from one or more mobile devices, and invocations of object methods redirected from the one or more mobile devices; to decode the encodings and update corresponding memory blocks on the cloud server; and to invoke and execute the object methods on the cloud server, and return execution results of the invoked object methods to the one or more mobile devices.
  • Example 55 may be example 54, wherein the cloud daemon may comprise a replication agent to receive compressively encoded memory blocks of memory from one or more mobile devices, encode corresponding replica memory blocks on the cloud server, determine and decode to recover updates to the replicate memory blocks, and apply the recovered updates to the replica memory blocks on the cloud server, continuously every t units of time.
  • the cloud daemon may comprise a replication agent to receive compressively encoded memory blocks of memory from one or more mobile devices, encode corresponding replica memory blocks on the cloud server, determine and decode to recover updates to the replicate memory blocks, and apply the recovered updates to the replica memory blocks on the cloud server, continuously every t units of time.
  • Example 56 may be example 55, wherein the replication agent may apply a sampling matrix ⁇ to a replica memory block to generate an encoding of the replica memory block.
  • Example 57 may be example 56, wherein the replication agent may apply a partial discrete cosine transform matrix ⁇ to the replica memory block to generate the encoding yi-i of the replica memory block.
  • Example 59 may be example 58, wherein the replication agent may further decode the compressively encoded update y ' to recover an update As to the replica memory block, and to apply the update As to the replica memory block.
  • Example 60 may be example 59, wherein the compressive encoding y, is compressed to reduce its size, and the replication agent may further decompress the compressed compressive encoding y t before calculating the compressively encoded update y '-
  • Example 61 may be one of examples 54-60, wherein the cloud daemon includes an object method servicer to receive invocations of object methods redirected from the one or more mobile devices; and to invoke and execute the object methods, and return execution results of the invoked object methods to the one or more mobile devices.
  • Example 62 may be example 61, wherein the object method servicer may translate an object pointer to a location in an address space of a mobile device to an object pointer to a location in an address space of the cloud server.
  • Example 63 may be example 61, wherein the object method servicer may serialize the execution results of the invoked object methods, and return the serialized execution results to the one or more mobile devices.
  • Example 64 may be a method for cloud computing, comprising: receiving, by a cloud server, encodings of memory blocks of memory compressively replicated from one or more mobile devices; decoding, by the cloud server, the encodings, and updating corresponding replica memory blocks of the cloud server; receiving, by the cloud server, invocations of object methods redirected from the one or more mobile devices; invoking and executing, by the cloud server, the object methods; and returning, by the cloud server, execution results of the invoked object methods to the one or more mobile devices.
  • Example 65 may be example 64, wherein receiving encodings may comprise receiving compressively encoded memory blocks of memory from one or more mobile devices; encoding corresponding replica memory blocks on the cloud server; determining and decoding to recover updates to the replicate memory blocks; and applying the recovered updates to the replica memory blocks on the cloud server, continuously every t units of time.
  • Example 66 may be example 65, wherein encoding corresponding replica memory blocks may comprise applying a sampling matrix ⁇ to a replica memory block to generate an encoding y ⁇ of the replica memory block.
  • Example 67 may be example 66, wherein applying may comprise applying a partial discrete cosine transform matrix ⁇ to the replica memory block to generate the encoding y ⁇ of the replica memory block.
  • Example 69 may be example 68, wherein decoding may comprise decoding the compressively encoded update y ' to recover an update As to the replica memory block.
  • Example 70 may be example 69, wherein the compressive encoding y t is compressed to reduce its size, and decoding further may comprise decompressing the compressed compressive encoding y t before calculating the compressively encoded update y '-
  • Example 71 may be example 64, wherein invoking may comprise translating an object pointer to a location in an address space of a mobile device to an object pointer to a location in an address space of the cloud server.
  • Example 72 may be any one of examples 64-71 , wherein returning may comprise serializing the execution results of the invoked object methods, and returning the serialized execution results to the one or more mobile devices.
  • Example 73 may be one or more computer-readable media having instructions stored thereon that cause a cloud server, in response to execution by the cloud server, to: receive encodings of memory blocks of memory compressively replicated from one or more mobile devices; decode the encodings and update corresponding replica memory blocks on the cloud server; receive invocations of object methods offloaded from the one or more mobile devices; and invoke and execute the object methods, and return execution results of the invoked object methods to the one or more mobile devices.
  • Example 74 may be example 73, wherein to receive encodings, to decode and to update may comprise to receive compressively encoded memory blocks of memory from one or more mobile devices, to encode corresponding replica memory blocks on the cloud server, to determine and decode to recover updates to the replicate memory blocks, and to apply the recovered updates to the replica memory blocks on the cloud server, continuously every t units of time.
  • Example 75 may be example 74, wherein to apply may comprise to apply a sampling matrix ⁇ to a replica memory block to generate an encoding y ⁇ of the replica memory block.
  • Example 76 may be example 75, wherein to apply may comprise to apply a partial discrete cosine transform matrix ⁇ to the replica memory block to generate the encoding y i of the replica memory block.
  • Example 78 may be example 77, wherein to decide may comprise to decode the compressively encoded update y ' to recover an update As to the replica memory block.
  • Example 79 may be example 78, wherein the compressive encoding y t is compressed to reduce its size, and to decode further comprise to decompress the compressed compressive encoding y t before calculating the compressively encoded update y '-
  • Example 80 may be example 73, wherein to receive invocations and to invoke and execute may comprise to translate an object pointer to a location in an address space of a mobile device to an object pointer to a location in an address space of the cloud server.
  • Example 81 may be any one of examples 73-80, wherein to return may comprise to serialize the execution results of the invoked object methods, and transmit the serialized execution results to the one or more mobile devices.
  • Example 82 may be a cloud server, comprising: one or more processors; memory coupled with the one or more processors; and means for receiving encodings of memory blocks of memory compressively replicated from one or more mobile devices; means for decoding, by the cloud server, the encodings, and updating corresponding replica memory blocks of the memory; means for receiving invocations of object methods redirected from the one or more mobile devices; means for invoking and executing, by the cloud server, the object methods; and means for returning execution results of the invoked object methods to the one or more mobile devices.
  • Example 83 may be example 82, wherein means for receiving encodings may comprise means for receiving compressively encoded memory blocks of memory from one or more mobile devices; means for encoding corresponding replica memory blocks on the cloud server; means for determining and decoding to recover updates to the replicate memory blocks; and means for applying the recovered updates to the replica memory blocks on the cloud server, continuously every t units of time.
  • Example 83 may be example 83, wherein means for encoding corresponding replica memory blocks may comprise means for applying a sampling matrix ⁇ to a replica memory block to generate an encoding of the replica memory block.
  • Example 85 may be example 84, wherein means for applying may comprise means for applying a partial discrete cosine transform matrix ⁇ to the replica memory block to generate the encoding of the replica memory block.
  • Example 87 may be example 86, wherein means for decoding may comprise means for decoding the compressively encoded update y ' to recover an update As to the replica memory block.
  • Example 88 may be example 87, wherein the compressive encoding y t is compressed to reduce its size, and means for decoding further may comprise means for decompressing the compressed compressive encoding y t before calculating the compressively encoded update y '.
  • Example 89 may be example 82, wherein means for invoking may comprise means for translating an object pointer to a location in an address space of a mobile device to an object pointer to a location in an address space of the cloud server.
  • Example 90 may be any one of examples 82-89, wherein means for returning may comprise means for serializing the execution results of the invoked object methods, and means for returning the serialized execution results to the one or more mobile devices.
  • One or more aspects or features of the subject matter described herein can be realized in digital electronic circuitry, integrated circuitry, specially designed application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) computer hardware, firmware, software, and/or combinations thereof.
  • ASICs application specific integrated circuits
  • FPGAs field programmable gate arrays
  • These various aspects or features can include implementation in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.
  • the programmable system or computing system may include clients and servers.
  • a client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other.
  • machine-readable medium refers to any computer program product, apparatus and/or device, such as for example magnetic discs, optical disks, memory, and Programmable Logic Devices (PLDs), used to provide machine instructions and/or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal.
  • machine-readable signal refers to any signal used to provide machine instructions and/or data to a programmable processor.
  • the machine-readable medium can store such machine instructions non-transitorily, such as for example as would a non-transient solid-state memory or a magnetic hard drive or any equivalent storage medium.
  • the machine-readable medium can alternatively, or additionally, store such machine instructions in a transient manner, such as for example, as would a processor cache or other random access memory associated with one or more physical processor cores.
  • one or more aspects or features of the subject matter described herein can be implemented on a computer having a display device, such as for example a cathode ray tube (CRT) or a liquid crystal display (LCD) or a light emitting diode (LED) monitor for displaying information to the user and a keyboard and a pointing device, such as for example a mouse or a trackball, by which the user may provide input to the computer.
  • a display device such as for example a cathode ray tube (CRT) or a liquid crystal display (LCD) or a light emitting diode (LED) monitor for displaying information to the user
  • LCD liquid crystal display
  • LED light emitting diode
  • a keyboard and a pointing device such as for example a mouse or a trackball
  • feedback provided to the user can be any form of sensory feedback, such as for example visual feedback, auditory feedback, or tactile feedback; and input from the user may be received in any form, including, but not limited to, acoustic, speech, or tactile input.
  • Other possible input devices include, but are not limited to, touch screens or other touch-sensitive devices such as single or multi-point resistive or capacitive track pads, voice recognition hardware and software, optical scanners, optical pointers, digital image capture devices and associated interpretation software, and the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Software Systems (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Stored Programmes (AREA)
  • Information Transfer Between Computers (AREA)

Abstract

L'invention concerne des appareils, des procédés et des supports de stockage associés à des aspects du délestage du traitement de dispositifs mobiles. Dans certains modes de réalisation, un dispositif informatique mobile peut comprendre un ou plusieurs processeurs ; une mémoire couplée auxdits un ou plusieurs processeurs ; et une couche de correction servant à répliquer par compression des blocs de mémoire de la mémoire dans un serveur en nuage, à délester par compression des appels de méthodes d'objet des objets résidant dans un bloc de mémoire de la mémoire vers le serveur en nuage, et à recevoir les résultats de l'exécution des méthodes d'objet appelées. D'autres modes de réalisation peuvent être décrits et/ou revendiqués.
PCT/US2015/019776 2014-03-10 2015-03-10 Accélération d'applications mobiles par l'intermédiaire d'un délestage à grain fin dans des infrastructures informatiques en nuage WO2015138504A1 (fr)

Priority Applications (3)

Application Number Priority Date Filing Date Title
CN201580008023.XA CN106462542B (zh) 2014-03-10 2015-03-10 通过向云计算基础设施的细粒度卸载进行的移动应用加速
CA2938697A CA2938697A1 (fr) 2014-03-10 2015-03-10 Acceleration d'applications mobiles par l'intermediaire d'un delestage a grain fin dans des infrastructures informatiques en nuage
US15/113,026 US10447813B2 (en) 2014-03-10 2015-03-10 Mobile application acceleration via fine-grain offloading to cloud computing infrastructures

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US201461950758P 2014-03-10 2014-03-10
US61/950,758 2014-03-10

Publications (2)

Publication Number Publication Date
WO2015138504A1 WO2015138504A1 (fr) 2015-09-17
WO2015138504A9 true WO2015138504A9 (fr) 2016-07-07

Family

ID=54072343

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2015/019776 WO2015138504A1 (fr) 2014-03-10 2015-03-10 Accélération d'applications mobiles par l'intermédiaire d'un délestage à grain fin dans des infrastructures informatiques en nuage

Country Status (4)

Country Link
US (1) US10447813B2 (fr)
CN (1) CN106462542B (fr)
CA (1) CA2938697A1 (fr)
WO (1) WO2015138504A1 (fr)

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3394751A1 (fr) * 2015-12-23 2018-10-31 IDAC Holdings, Inc. Procédés de délestage d'un calcul d'un dispositif mobile à un nuage
CN106507131B (zh) * 2016-12-07 2019-04-30 清华大学深圳研究生院 一种超高清视频编码的自适应云卸载方法
US11204808B2 (en) * 2016-12-12 2021-12-21 Intel Corporation Offload computing protocol
US10440096B2 (en) * 2016-12-28 2019-10-08 Intel IP Corporation Application computation offloading for mobile edge computing
US11468542B2 (en) * 2019-01-18 2022-10-11 Arizona Board Of Regents On Behalf Of Arizona State University LAPRAN: a scalable Laplacian pyramid reconstructive adversarial network for flexible compressive sensing reconstruction
US11294702B2 (en) 2019-05-01 2022-04-05 EMC IP Holding Company LLC Method and system for processing data using a processing pipeline and processing units
US11119803B2 (en) * 2019-05-01 2021-09-14 EMC IP Holding Company LLC Method and system for offloading parity processing
CN110377352B (zh) * 2019-05-20 2022-06-14 中国人民解放军军事科学院战争研究院 基于移动设备云系统的任务处理方法和装置
US20220405104A1 (en) * 2021-06-22 2022-12-22 Vmware, Inc. Cross platform and platform agnostic accelerator remoting service

Family Cites Families (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
FR1577574A (fr) * 1967-10-30 1969-08-08
CN1232914C (zh) * 2000-04-26 2005-12-21 计算机合作者思维公司 在分布计算机系统中保持数据完整性的方法与设备
US6990662B2 (en) * 2001-10-31 2006-01-24 Hewlett-Packard Development Company, L.P. Method and system for offloading execution and resources for resource-constrained networked devices
US7072807B2 (en) * 2003-03-06 2006-07-04 Microsoft Corporation Architecture for distributed computing system and automated design, deployment, and management of distributed applications
US7409709B2 (en) * 2005-02-14 2008-08-05 Etsec, Inc. Systems and methods for automatically reconfiguring a network device
DE602006014214D1 (de) * 2006-03-03 2010-06-17 Europ Agence Spatiale Verfahren zum verarbeiten von positionsbestimmungssignalen insbesondere für innenanwendungen
US8194088B1 (en) * 2006-08-03 2012-06-05 Apple Inc. Selective composite rendering
US20080219579A1 (en) * 2007-03-05 2008-09-11 Aksyuk Vladimir A Methods and Apparatus for Compressed Imaging Using Modulation in Pupil Plane
US8074014B2 (en) * 2008-03-31 2011-12-06 Microsoft Corporation Storage systems using write off-loading
US20110117956A1 (en) * 2009-11-17 2011-05-19 Yosi Levi Industrial radio device with unified programming interface and methods
US8352776B2 (en) * 2010-04-02 2013-01-08 Cisco Technology, Inc. Facilitating persistence of routing states
CA2737822C (fr) * 2010-08-31 2019-02-19 Mirza F. Beg Systeme et procede d'acquisition rapide d'images de tomographie par coherence optique (oct) au moyen d'un echantillonnage compressif
US8489760B2 (en) * 2011-03-31 2013-07-16 Juniper Networks, Inc. Media file storage format and adaptive delivery system
US9084036B2 (en) * 2011-05-17 2015-07-14 Alcatel Lucent Signal source localization using compressive measurements
US9712599B2 (en) 2011-10-03 2017-07-18 International Business Machines Corporation Application peak load processing
CN102629198B (zh) * 2012-02-29 2015-05-13 北京大学 软件应用中计算逻辑的按需远程执行方法
US8831121B1 (en) * 2012-06-08 2014-09-09 Vt Idirect, Inc. Multicarrier channelization and demodulation apparatus and method
US20140337408A1 (en) * 2013-05-10 2014-11-13 Appsense Limited Systems, methods and media for minimizing data downloads

Also Published As

Publication number Publication date
WO2015138504A1 (fr) 2015-09-17
US20160344844A1 (en) 2016-11-24
CA2938697A1 (fr) 2015-09-17
US10447813B2 (en) 2019-10-15
CN106462542B (zh) 2020-03-17
CN106462542A (zh) 2017-02-22

Similar Documents

Publication Publication Date Title
US10447813B2 (en) Mobile application acceleration via fine-grain offloading to cloud computing infrastructures
US9740402B2 (en) Migrating workloads across host computing systems based on remote cache content usage characteristics
JP2022501736A (ja) オンデマンドコード実行システムにおける実行環境についての効率的な状態メンテナンス
US10552274B1 (en) Asynchronous in-memory data checkpointing for distributed computing systems
US20130346465A1 (en) Application enhancement using edge data center
US11018692B2 (en) Floating point data set compression
US11316530B2 (en) Adaptive compression for data services
US20150261459A1 (en) Migrating workloads across host computing systems based on cache content usage characteristics
US10701172B2 (en) Clients for storage services
US20170193028A1 (en) Delta encoding in storage clients
US20190007483A1 (en) Server architecture having dedicated compute resources for processing infrastructure-related workloads
US9940245B2 (en) Caching in storage clients
Valery et al. Low precision deep learning training on mobile heterogeneous platform
CN111767059B (zh) 深度学习模型的部署方法、装置、电子设备和存储介质
EP3754499A1 (fr) Génération de modèles de configuration pour la commande de distribution d'applications
US10198262B2 (en) Adaptive core grouping
US10637915B1 (en) Storage services configured for storage-oriented applications
US20230128877A1 (en) Sharded Database Load Distributor
US11196817B1 (en) Intelligently managing resource utilization in desktop virtualization environments
US20230396687A1 (en) Application service behavior management using request context
US11056112B2 (en) Voice-based operations

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 15762137

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 15113026

Country of ref document: US

ENP Entry into the national phase

Ref document number: 2938697

Country of ref document: CA

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 15762137

Country of ref document: EP

Kind code of ref document: A1